Skip to main content

Full text of "Effect of tobacco advertising on tobacco consumption : a discussion document reviewing the evidence"

See other formats


EFFECT OF TOBACCO ADVERTISING ON TOBACCO CONSUMPTION: A 

DISCUSSION DOCUMENT REVIEWING THE EVIDENCE 

This report reviews the evidence on the effects of tobacco advertising on tobacco 
consumption, including the effect of advertising bans. The report draws on the 
available published material in this field, both from the UK and internationally, 
together with some original econometric work. 

The report is being published as a discussion document. The Department of 
Health's Economics and Operational Research Division would welcome any 
comments on the overall analysis, methodology and original pieces of work in the 
report. Any comments should be received by Friday 29 January 1993, and should 
be addressed to: 



Mr K Leaney 

Economics and Operational Research Division 

Department of Health 

Room 2813 

Millbank Tower 

21-24 Millbank 

LONDON SW1P4QU 



- i - 







-r / 



Printed image digitised by the University of Southampton Library Digitisation Unit 



ACKNOWLEDGMENTS 



This study was carried out by economists in the Economics and Operational 
Research Division of the Department of Health under the supervision of Clive Smee 
and Michael Parsonage. Robert Anderson was primarily responsible for drafting the 
main body of the report and Annex A. Simeon Duckworth drafted Annex B as well 
as undertaking the analysis which it reports. 

The study benefited from comments made by medical and administrative 
colleagues within the Department of Health and by economists in the Treasury; 
Customs and Excise; and the Department of Trade and Industry. 

Thanks are due to the Tobacco Advisory Council and Action on Smoking and 
Health for contributing material and analysis which strengthened the body of 
evidence on which the report is based. 



Clive Smee 

Chief Economic Adviser 



October 1992 




Printed image digitised by the University of Southampton Library Digitisation Unit 



CONTENTS 



Page 



Introduction 1 

Indirect and circumstantial evidence 2 

Analysis of incentive structure in tobacco market 2 

Mechanisms by which advertising might increase 

consumption 4 

Evidence on reactions to advertisements 5 

Evidence from surveys of reasons for starting 

smoking 6 

Other evidence of indirect effects 7 

Quantitative direct evidence 7 

Cross-section analysis of countries with 

different levels of controls 1 0 

Time series analysis of fluctuations in 

advertising expenditure within countries 1 2 

Before-and-after studies of consumption in 

banning countries 16 

Norway 17 

Finland 18 

Canada 19 

New Zealand 20 

Summary and Conclusions 21 

Annex A: Time series analysis of fluctuations in 

advertising expenditure within countries: further 

analysis 23 

United Kingdom 23 

United States 29 

Other countries 33 




Printed image digitised by the University of Southampton Library Digitisation Unit 



Annex B: Analysis of tobacco consumption in Norway 
and the UK; Multicollinearity 



35 



Norway 35 

United Kingdom 40 

Prevalence 47 

Multicollinearity 48 

References 51 



- iv - 

Printed image digitised by the University of Southampton Library Digitisation Unit 



INTRODUCTION 



1 . There has been much dispute about the influence of tobacco advertising on 
smoking. The issue has recently been the subject of official reviews in the United 
States 1 , New Zealand 2 and Canada 3 . Official reviews are currently under way 
In Sweden and Germany. Chapman (1986) and Godfrey (1990) review the 
literature. This paper examines the evidence on the effect of tobacco advertising, 
including the effect of advertising bans. 

2. There is a wide range of studies in this area following a variety of methods. 
The studies cover knowledge of advertisements, reasons for starting to smoke, 
opinions as to the effect of advertising on smoking behaviour and empirical work 
looking at associations between smoking behaviour and advertising expenditure 
over time or across countries. The Report of the US Surgeon General shows the 
range of studies and gives references. The forms of evidence clearly differ in 
nature. We have looked at the evidence from abroad as well as in this country and 
we have carried out some analysis of our own where published studies do not 
exist. We have concentrated attention on studies which follow sound methods 
with a reasonable prospect of delivering reliable answers. The evidence falls into 
two broad categories: 



Indirect and Circumstantial Evidence 

(a) analysis of the incentive structure in the tobacco market to determine 
whether It is likely that advertising will be undertaken to increase overall 
consumption; 

(b) review of mechanisms by which advertising might increase consumption; 

(c) evidence on reactions to advertisements, particularly among young 
people, on the assumption that there has to be knowledge of advertising if 
it is to have any effect on smoking; 

(d) surveys of reasons for taking up smoking; and 

(e) evidence on the indirect effects of advertising. 



1 Report by the US Surgeon General Reducing the Health Consequences of 
Smoking: US DHHS (1989). 

2 Report by the Toxic Substances Board (NZ Department of Health 1989). 

3 Appeal against the Tobacco Products Control Act (TPCA). 

- 1 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



Quantitative Direct Evidence 



Statistical analysis of smoking behaviour or tobacco consumption in the face 
of: 

(a) variations in the amount of advertising; 

(b) variations in the level of controls on advertising; and 

(c) a ban on advertising. 



INDIRECT AND CIRCUMSTANTIAL EVIDENCE 
Analysis of Incentive Structure in Tobacco Market 

3. The first step is to analyse the characteristics of the tobacco market to assess 
whether there exists an incentive a priori to undertake advertising with the aim of 
increasing overall consumption. 

4. It could prove profitable for a monopolistic firm to advertise in order to expand 
the market (strictly, move the demand curve outwards). The amount worth 
spending depends on the cost and effectiveness of advertising. 

5. In an industry with more than one supplier it is much less likely to be 
worthwhile for one firm to advertise to expand the market because it only reaps 
a fraction of the benefit. 

6. It would pay the firms in any industry to collude to achieve the effects of a 
monopoly, including advertising to expand industry sales. The tobacco industry in 
this country is highly concentrated as figures summarised by Booth et a! (1990) 
show. Four firms account for 99% of domestic production and 85%-90% of sales. 
The largest single supplier to the domestic market has 44% of sales. While some 
of the pre-conditions for successful collusion are present in the tobacco industry, 
there is no evidence of collusive arrangements to promote a monopoly. 

7. However, firms create a degree of monopoly for their own product by branding 
supported by advertising and of course firms advertise brands not generic tobacco 
products. The idea of brand loyalty merely expresses the monopoly power of a 
brand. It may therefore be profitable for a firm to advertise to encourage the 
uptake of smoking using the advertised brand, since in that case the firm captures 
all the benefit of the increase in industry sales due to its advertising. 

8. The industry states that the sole purpose of advertising is for each firm to 
maintain and if possible increase market share. Analysis carried out within the 
industry confirms that advertising does indeed affect brand shares. However, it 
is quite possible that some people are recruited to smoking by brand advertising 
even if firms do not specifically set out to attract them. As we shall see below 

- 2 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



most of these are likely to be teenagers. 



9. It is worth looking at the nature of the product and smoking careers for the 
incentives they create for advertisers. Smoking is not a necessity of life, so that 
recruitment of new smokers is necessary to maintain the smoking population by 
replacing those who die or give up. Advertising may therefore have a role in 
promoting recruitment to this non-necessary product. Smokers typically start in 
their teens. The OPCS survey of smoking among schoolchildren (OPCS (1989)) 
and the 1988 General Household Survey (OPCS (1990a)) show the following 
pattern of smoking prevalence with age: 



OCCASIONAL AND REGULAR SMOKERS 1 1988 





% 




Age 


Males 


Fern; 


11-12 


4 


1 


13 


10 


9 


14 


15 


19 


15 


24 


31 


16-19 


28 


28 



In 1 988 84% of smokers aged 1 6 or over had become regular smokers before their 
twentieth birthday; 45% of male and 39% of female smokers in the 16-34 age 
group started before the age of 1 6. 

10. About two thirds of men who have ever smoked regularly have given up by 
the age of 60 (GHS 1988 table 5.3). Nevertheless, the typical smoking career 
lasts for many years. Applying current life-table methods and taking account of 
the much higher death rate among smokers from 35 onwards, the typical length 
of a smoking career for a man of twenty who smokes regularly is just over twenty- 
six years. And of course about half of twenty-year-old smokers already have a 
five-year smoking career behind them. The average smoking career for a woman 
of twenty is rather higher at about thirty-three years 2 . In consequence, 



1 age 11-15: occasional (current smokers usually smoking less than one cigarette 
a week) or regular (one or more cigarettes a week) smokers. 

age 16-19: answer of yes to "Do you smoke cigarettes at all nowadays?" 

2 Relative death rates for smokers are taken from the Cancer Prevention Study II, 
reported in the Surgeon General's Report US DHHS (1989), pp150-151. The 
expected smoking career of twenty year old women is longer than men for three 
reasons: (a) the relative mortality rates of female smokers to female non-smokers 
are lower than for males in CPS II; (b) women have lower mortality rates than men; 
and (c) the rate of smoking cessation among women by age 60 is lower for men: 
54% against 66%. 

-3- 

i 

! 

Printed image digitised by the University of Southampton Library Digitisation Unit 



advertising to increase smoking prevalence, particularly among young people, has 
the potential to deliver something like thirty years of consumption from each new 
consumer. 

11. To some, the industry's opposition to a ban suggests that advertising does 
increase safes, if advertising does not increase sales, then the industry collectively 
loses nothing from a ban and it gains the £1G0m or so it spends each year on 
advertising. There are strong counter-arguments. To begin with, there is the 
obvious point that advertising protects the branded home product against generic 
imports. Secondly, advertising may offer the least costly way of increasing or 
maintaining market share. Thirdly, advertising allows the maintenance of brands 
with (real or imagined) characteristics which command a price premium over a 

generic product. In this way ail companies could gain from advertisinc even if 
there occurred no increase in total sales. 



Adverti sing Might Increase Consumption 

1 2. it may be worth bearing in mind the direct mechanisms by which advertising 

fnc Z°!rT increase tobacco consumption (US Surgeon General's Report 

iu& DnHb {1989)): 



- by inducing children and 
tobacco products and in this 



young people to begin experimenting with 
way initiate regular smoking 



- by encouraging adults to take up smokincj 



“ by enc °uraging existing smokers to smoke more 



- by undermining existing smokers' motivation to give up 
~ by erscoura gi n g former smokers to resume the habit. 



Advertising may also increase smoking indirectly: 

d *" 3 ™ •»“« « - h “" 

harmful effeMs°y D mb C accran^ tl in n Miir aV resl,airi thair discussion of the 
education messages. way res trict the flow of health 



- 4 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



Evidence on Reactions, to Advertisements 



13. There is great concern about awareness of tobacco advertisements among 
children and young people, because many (45% of male and 39% of female 
smokers in the 1 6-34 age group) start smoking before the age of 16, that is, 
before the age at which traders are allowed to sell them tobacco. In other words, 
they start when by implication they are not old enough to take decisions about 
smoking for themselves. There is also concern that advertisers may be targeting 
these age groups to recruit new smokers and doing it successfully. The evidence 
about awareness is beyond dispute, for example, Aitken et al (1987) in the UK 
have shown that most primary school children are aware of tobacco advertising 
and many can supply the brand names of cigarette advertisements from which this 
information has been removed. There is even evidence of awareness of cigarette 
brands among pre-schooS children (Fischer et al (1991)). 

14. There is some circumstantial evidence that advertising may increase smoking 
prevalence particularly among children: findings from a number of countries 
indicate that by and large smokers and those who later take up smoking find 
tobacco advertising more attractive and have a more positive attitude towards it 
than non-smokers (Vickers (1992)). 

15. The Camel advertising campaign in the US using the Old Joe character has 
aroused particular concern on the score of its appeal to young people, its influence 
on their smoking patterns and what it reveals about the tobacco companies' 
advertising strategy and capabilities. An article by DiFranza et al (1991) shows 
that a high proportion of teenagers recognise the character, higher than among 
adults. This campaign provides an example of the success of advertising in 
capturing brand share. It has taken Camel from nowhere in the under age smoking 
market to a one third share. It has unquestionably achieved greater success among 
children than among adults. The paper quotes extensive evidence of tobacco 
companies' approach to marketing, some of it from the TPCA appeal in Canada, 
which suggests that campaigns of this kind are typically aimed at capturing the 
early teenage market and may also be intended to encourage more people in this 
age group to take up smoking. However, the evidence on consumption is not 
sufficient to establish that the campaign does actually increase smoking; it merely 
relates to market share. 

1 6. It is worth noting that in the UK a campaign of this kind making an appeal to 
children, whether intentionally or not, would not be allowed under the voluntary 
agreements between the government and the tobacco industry. There is 
nevertheless some indication that advertising influences market share among 
smokers below the age at which they can legally be sold cigarettes in the UK. A 
survey by Roberts (1990) shows that brand shares among 11 to 14 year olds are 
highest for the products which are most heavily advertised. 

17. A survey from California reported by Pierce et al (1991) confirms DiFranza's 
findings about the effect of advertising on young people. It shows that perception 
of advertising is higher among young smokers (defined as 12-18) than adults; 

- 5 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



market share patterns over a four-year period followed perceived advertising 
patterns; and changes in market share resulting from advertising occur mainly in 
young smokers. Again, however, the results show that advertising influences 
market share; the evidence provides no information about the influences on the 
overall level of tobacco consumption among young people. 



■s ij, Aitken ©t al { t argue that the positive attitudes smokers have towards 
tobacco advertising indicate that they are deriving some benefit from advertising, 
probably perceiving it as a form of social approval of smoking. The fact that those 
who later become smokers share these views suggests that tobacco advertising 
may be playing some part in breaking down the resistance to smoking based on 
awareness of the health risks. Tobacco sponsorship of sport in particular may 
cultivate these positive attitudes by associating tobacco with characteristics which 
young people admire. Aitken 's survey suggests that the sport-sponsoring brands 
of cigarette are indeed associated in young people's minds with excitino sports 
{though not necessarily the sport sponsored). However, there was no difference 
between smokers and non-smokers in their awareness of brand sponsorship. 



19 ‘ A , r « cent 0PCS (1990b) survey of secondary school children has uncovered 
H tors associated with starting to smoke. These include family structure 
educations aspiration, attitudes of parents and siblings, and a reladvelj positive 
Ikkk “ vvards smok ' n 9- The survey interviewed a panel at the beginning of the 
o detmine a w ° e r h ^ ° f SCho ° L This lo "9^"al feature made ?o££e 



children in Airctralial ‘pL^Xourabtto C ° h °' tSt ^ of 10-12 year old 
of an initial survey were uncovered ? ^ adoptlon of smoking within a year 

the order of importance in the ™ ° 9 ' St!0 regression. This technique allows 

contribution of each factor '■ r,tnbuto y (actors to emerge and quantifies the 
factors and accounts for about 7% n°r th ° f ad k ertisif19 c °™s fourth out of five 
factor, unhelpfully, is age a Hme , f thfT J" 9 Vanation ex P |ai "ed. The leading 
smoking prevalence among siblings and peers r S ® COnd and third 
not play a part. Those stopping smokina dur’ina rhe^ ^ parental smokin 3 do 
pattern, with disapproval of advertising coming third in the Tst^ ° PP ° Site 

fheir HmSnsTs^ttere 8 ^ ° PCS (1 " 0b) stud V hav e 

behaviour could be linked, the assr 't ' advertising to which smoking 

- me associations uncovered represent circumstantial 

- 6 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



evidence that advertising increases tobacco consumption, it remains possible that 
the causation runs in the other direction - children disposed to smoke are more 
likely to react positively to tobacco advertising and show greater awareness of it. 



Other Evidence of indirect Effects 

22. Tobacco advertising may encourage smoking by providing smokers with an 
excuse to play down the danger to their health. A survey by Marsh et a! (1983) 
shows that despite the Government's own health education messages 44% of 
smokers (but only 26% of non-smokers) agreed with the statement that "Smoking 
can't be really dangerous or the Government would ban cigarette advertising." The 
implication is that a ban would itself send a powerful health message. 

23. Evidence from the US suggests that magazines restrict their coverage of the 
dangers of smoking because they fear a loss of revenue from tobacco advertising. 
A study by Warner et al (1992) looked at a wide range of magazines over the 
period 1 959-86 and using logistic regression related the probability of publishing 
an article on the health risks of smoking to acceptance of tobacco advertising in 
the same year. The effect was strongly established for women's magazines in 
particular, even after taking account of possible confounding variables (such as size 
of readership and coverage of health issues generally). For example, the probability 
of covering smoking dangers was 38% lower in magazines with average reliance 
on tobacco advertising compared with magazines not accepting tobacco 
advertising. There remains the problem of the direction of causation. For example, 
it may be that a magazine's attitude towards smoking determines whether it will 
accept tobacco advertising. However, coverage of the health risks of smoking was 
related to the proportion of advertising revenues derived from tobacco advertising, 
not merely whether it was accepted or not, and this tends to suggest that some 
magazines have modified their stance in deference to tobacco advertisers. 



QUANTITATIVE DIRECT EVIDENCE 

24. It is not possible to mount a controlled experiment to examine the effect of 
advertising on consumption, with one group exposed to advertising and a control 
group not exposed. However, a number of situations have arisen where variations 
in advertising have occurred which make it possible to investigate the effects of 
exposure on consumption. Such "natural experiments" fall into three broad 
categories: 



(i) different levels of advertising controls seen in a cross-section of 
countries; 

(ii) year-to-year fluctuations in advertising expenditure within countries; 

(iii) the period before and after the introduction of advertising bans in 

- 7 - 

D rinted image digitised by the University of Southampton Library Digitisation Unit 



certain countries. 



25. Advertising is only one of many potential influences on smoking in these 
contexts, and it is important to control as far as possible for the influence of other 
variables to avoid "confounding” - attributing the effects of other variables to 
advertising. Most studies in these contexts use multivariate regression for this 
purpose; those which do not are less reliable. We do not report studies whose 
methods have obvious defects, such as those which note that smoking declined 
following a ban and conclude that the ban caused the decline without adequate 
examination of either trends before the ban or of other influences such as price 
rises which accompanied a ban. 

26. Regression methods nevertheless encounter a range of technical problems, 
some of which are more important in one experimental context than another. 

27. Some of these problems relate to the specification of the relationship, others 
to weaknesses in the data, though the two sets of problems interact. We first 
consider specification problems. The specification of a regression relationship 
refers to its mathematical form and the variables it incorporates. 



’ L he forn ? . chosen for * he re,ation ship inevitably imposes some restriction on 
the effects which can be picked up. For example, a specification which expects 
a ban to usher in an immediate and sustained reduction in smokina will deliver an 



w. v.w. s , a kj , , ^uusumption 




may last for many years, by inducing a 



- 8 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



teenager to start smoking, for example. This feature indicates a need to include 
lagged consumption among the explanatory variables to enable the full long term 
effect of the other explanatory variables to be uncovered. 

32. The data base for the regression may not be ideal. Potential explanatory 
variables often move together over time and this makes it difficult to distinguish 
their separate effects (multicoliinearity). The data may contain measurement 
errors. It is sometimes incomplete - for example, the published data on advertising 
expenditure in the UK omits billboard advertising. 

33. These problems affect the various experimental contexts differently. 
International cross-sectional studies tend to raise the suspicion that omitted 
variables such as social attitudes towards smoking play a part in low smoking 
levels and in creation of a social environment favouring a ban. 

34. The bulk of the published material studies the effect of year-to-year 
fluctuations in advertising expenditures within countries. This experimental 
context has its drawbacks as a guide to the effect of a ban: 



- the effect of advertising is likely to be small in comparison with price and 
income and is therefore more likely to fail statistical tests due to the 
imprecision of estimates; 

- the annual fluctuations presumably relate to the least productive slice of 

expenditure, on the reasonable assumption that advertising is subject to 
diminishing returns. Accordingly, the effect of an outright ban which cuts 
away the most effective core of advertising is likely to be greater than 
extrapolation would suggest. The US Surgeon General's Report (US 
DHHS (1989)) presents a strong if somewhat speculative version of this 
argument: since the total level of advertising expenditure which is 

worthwhile to contest market share lies well above the level associated with 
zero marginal impact on total tobacco consumption, a full ban could have an 
effect even if the evidence based on year-to-year fluctuations points to no 
effect. 



35. A total ban is a much more promising experimental context which gets round 
many of the difficulties inherent in the study of year-to-year fluctuations, 
particularly causation running in both directions. Moreover, if there is an effect it 
will be on a larger scale and should show up more clearly. It is worth bearing in 
mind that the effect may also be enhanced by the accompanying publicity and by 
the implicit message that government now believes that the dangers of smoking 
justify a ban. 

36. Predicting the effect of a ban has to rely to some degree on evidence from 
other countries. A number of factors need to be borne in mind in transferring 
results between countries: 

-9- 



Printed image digitised by the University of Southampton Library Digitisation Unit 



- in predicting the effects of a ban it is important to take account of the 
degree of restriction already in place. Finland, Norway and Canada imposed 
bans from a much lower base of restriction on advertising than New 
Zealand. The UK/s current level of restriction lies closer to the pre-ban level 
in New Zealand than to pre-ban levels in the other three countries (Laugesen 
(1991)); 

- in some countries various other measures accompanied the introduction of 
a ban and these are likely to have had some effect on smoking; 

because of cultural factors, tobacco consumption may not react to 
changes in price and advertising variables in quite the same way in different 
countries; 

- the degree of market power in the tobacco market may vary from country 

to country and with it the proportion of advertising undertaken with a 
market share objective. 



s 



Cross-section „ Analysis o f _Co_un tries. with Different Levels of Controls 
37. i here are two international cross-sectional studies. 

• , A „ studv bv Cox et al n 984) makes an attempt to assess the effectiveness of 
the different approaches to tobacco advertising control policy seen in different 
countries. The study identifies two basic philosophies: a legislative approach and 
a piecemeal approach based on voluntary agreements. In the absence of data on 

SXpenditur f in the different countries, the study resorts to an indirect 
f assessing the effect of control policy on smoking. The method is to 

TZIT* f r ' es rS9reSSion over a similar time period for each country using 
timp trp °" spec ' f ' Cdtlon ' explaining tobacco consumption by price, income and a 

“^r nt that tobacco control policy is influencing consumption 
its effect should show up in the regression equation in two ways. Firstly the 
p oportion of the variation in tobacco consumption explained by the regression 
should be lower in the countries where the control policy is more effective^ 
nfluencmg consumption. Secondly, all the estimators correlated with the omitted 
vanable will be biased and inefficient, but the authors focus on the effect on the 
ime trend, arguing that it should plot a steeper downward course in the countries 
which nave achieved more effective control as it picks up the long term effects on 

*° p !, p c ° s° nsum Pt'°n- These effects are indeed found and the authors conclude 
that legislative regimes work better than voluntary arrangements. However this 
evidence is suggestive rather than conclusive. Other factors could account for the 
S" „ results . S °* a < trends away from smoking may be stronge in Ee 

control of" ^“adverSsi^ 0 ' 63 ' 6 '' 3 C " mate ° f ° Pini ° n fav0urable to legislative 



- 10 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



39. laugesen et al (1991) have carried out a multivariate pooled cross- 
section/time series international study using a sample of annual data covering 22 
OECD countries over the period 1960-1986 . The explanatory variables include 
price, GDP per head, a score on a scale of 0-10 to capture the level of advertising 
restrictions and certain other variables such as the proportion of tobacco accounted 
for by manufactured cigarettes. The advertising coefficient suggests that for each 
point on the advertising score consumption falls by 0,6%. In a second 
specification the coefficients are allowed to vary over time by combining dummy 
variables representing succeeding years with the year's values of the explanatory 
variables. The advertising variable registered a perverse positive sign in the early 
years but it became negative in the early 1 970s and has grown steadily in value 
since then. There is no obvious explanation for the unexpected positive sign in the 
early period. The coefficient for the latest year suggests that each point on the 
advertising score would reduce consumption by 1%% (0.6% where the 
coefficients are held constant over time). The UK's advertising score was 6.0 (out 
of 10) in 1986, though subsequent agreements on sports sponsorship and 
advertising will have increased it marginally. 

40. The system for calculating a country's tobacco control score is set out in 
Appendix 4 of Health or Tobacco (NZ Department of Health (1989)). A number of 
features of the weighting system are open to question. For example, a ban on 
television advertising gets the same score as a ban on cinema advertising, despite 
the very much higher audiences for television. It would be interesting to see the 
effect of different scoring systems. 

41 . The main difficulty with this experimental context is uncertainty about the 
direction of causation and possible confounding factors: negative social attitudes 
towards smoking are likely to lead to low tobacco consumption and to strict 
controls on tobacco advertising. In these circumstances a cross-section study 
would suggest that advertising restrictions are responsible for low tobacco 
consumption whereas in reality a third factor causes both. 



1 Stewart (1991) has cast doubt on the reliability of some of the data used in this 
study with all the variables affected in some degree. He argues that errors in the 
data discredit Laugesen's results. However, the effects of data error are not as 
serious as he suggests. The effect of "white noise" measurement error on the 
advertising coefficient is clear in the case of the dependent variable and the 
advertising variable, but not in the case of the other explanatory variables (Greene 
(1991a)). Measurement error in the dependent variable increases the variance of 
coefficients but does not bias their values. Measurement error in the advertising 
variable leads to "attenuation" biasing its own coefficient towards zero. 
Accordingly, errors in the advertising variable may not be critical. However, 
Stewart also makes persuasive recommendations as to better ways of measuring 
some of the explanatory variables. Errors in these variables impart bias to the 
advertising coefficient but the direction of bias is not in general known. Stewart 
does not offer a fresh analysis on a better data set. In the meantime it would be 
wrong to reject Laugesen's results on the score of imperfect data. 




y 1 - 



Printed image digitised by the University of Southampton Library Digitisation Unit 



42. Tiie bulk of the published material studies the effect of year-to-year 
fluctuations in advertising expenditures within countries, mainly this country or the 
ted States. Ail the studies reported here take account of other influences on 
a* onsumption to avoid confounding, not only price and income variables 
but also specific health "scares" such as reports of the Royal College of Physicians. 
Many of these articles are of very high quality using state-of-the-art econometric 
technique. Advertising is not always the only or even the main focus of interest. 
Many have found that advertising has an effect; some have not. There appears to 
be a tendency for articles which have found an effect to underestimate its scale. 

e [ e f sor ! is that results are usually expressed in terms of an elasticity fi gure 
which implicitly invites comparison between different variables standardised on a 
1 /o cll ange. However, in order to standardise the evidence with that on 
advertising bans, it is more useful to show the effect using a 100% change. The 
results presented in this section follow this method, though as we saw in 
paragraph 34 there are good arguments for thinking the effect of a ban would be 
greater than simple extrapolation would indicate. 

4 J ' S p ® c,al factors are at work in different countries. Independent researchers in 
the UK have to work within the limitations imposed by the incomplete advertising 
data re,eased b Y the tobacco industry, covering broadcast media and the press but 
omitting billboard advertising and sponsorship. The effect of this restriction 
tiecame more serious following the ban on cigarette advertising on television in 
1965 ‘ Radfar (1985) estimates that in the mid-1980s poster advertising 
accounted for 30%-40% of expenditure on cigarette advertising. The Metra 
consulting group commissioned by the tobacco industry had access to the full 
advertising data apart from sponsorship which had not then, in 1 979, attained the 
importance it appears to have today. Studies carried out in the United States are 
particularly informative, because comprehensive data on advertising expenditure 
are freely available to independent researchers. 

44. The studies differ in specification, eg the presence or absence of a lag on 

consumption or a carryover effect of advertising from one time period to the next. 

Most studies have to rely exclusively on time series data, in the US, however it 

is possible to use pooled data taking advantage of cross-sectional variation in price 
across states. H 

45. The results of the key published studies are summarised in the table overleaf, 
bmce the full effects of a reduction in advertising may take some time to come 
through, some of the studies have provided short run and long run estimates 
(a owing for a habit effect). A range is shown where studies report short and long 
term effects or results from different specifications. The table also shows the 
publication dates of the studies and the period covered by the data they have used. 



- 12 - 



Printed image digitised by the University of Southampton Library Digitisation Unit